Different floating point operation results on AMD and Intel

ykwok · 06-15-2008, 09:21 PM

Hi,

I run an application which does floating point computations on an AMD Althon X2 and Intel P4. Surprisingly, I obtained different results even though I used the same compiled binary and both machines are running Linux.

Any idea?

Thank you!

pinniped · 06-15-2008, 09:44 PM

Can you post a minimal program that shows the difference in results?

I have encountered similar problems when compiling F77 code, but I never had time to look into it since it was someone else's code and I had no clue where to start.

We might have to get our hands dirty and possibly even go so far as to write some assembler code to find out what's going wrong (just to check whether it's a compiler/library bug or a FPU bug).

ykwok · 06-15-2008, 10:00 PM

Hi pinniped,

Thank you very much for your suggestion.

Unfortunately, the application is not free and is copyrighted. I cannot post the code here.

The suggestion of examining the asm code is good.

Thanks again!

chrism01 · 06-16-2008, 08:52 PM

Note pinniped's 1st comment: "Can you post a minimal program that shows the difference in results?"
We don't need to see your prod code, we just need a minimal (might only be 5 or 10 lines) of a prog you make up just to demo this issue. Feel free to use var names like v1, v2, etc.

ykwok · 06-17-2008, 06:56 AM

Hi Chris,

Thanks for writing.

It's a big code. we are now still trying to pin down which part of the application causes the problem.

Thank you again!